Skip to content

Export Linux audit status as Prometheus metrics

License

Notifications You must be signed in to change notification settings

TeachersPayTeachers/linux-audit-exporter

Repository files navigation

linux-audit-exporter

A Prometheus exporter for Linux Audit status.

Project Status

This project is:

  • In ALPHA state
  • Is maintained by Teachers Pay Teachers
  • Is used in production by Teachers Pay Teachers

Versioning

While this project is in ALPHA status, breaking changes may occur between minor versions, and will be announced in CHANGELOG.md and in the GitHub release notes.

The API surface area includes:

  • Runtime dependencies
  • The Helm chart values
  • The CLI flags and environment variables
  • The exported Prometheus metric names and labels

Requirements

Golang 1.14+ is required to build the project.

Will run on any system that can run Go binaries, but will only export Linux Audit status when run from a Linux host. When run from a non-Linux host, it will export zero-ed metrics.

Additionally, requires:

  • Root (UID 0) user to run binary
  • Audit read privilege (CAP_AUDIT_READ)
  • --host=pid in Docker, hostPID: true in Kubernetes

Why

The Linux audit system can be configured to log security-relevant events such as syscalls, and processed by auditd or commercial alternatives.

On high-throughput systems it may generate audit records at a faster pace than auditd (or equivalent) can consume. Depending on how Linux audit is configured, this can result in:

  • Lost audit events
  • Backlog processing delays
  • Out-of-memory

The linux-audit-exporter surfaces metrics meant to help operators monitor the status of the Linux audit system, and avoid these outcomes.

Build

Go binary

$ go build .

Docker image

$ docker build .

Usage

Help

$ linux-audit-exporter -h
Export Linux audit status as Prometheus metrics

Usage:
  linux-audit-exporter [flags]

Flags:
      --health-path string      health path (default "/healthz")
  -h, --help                    help for linux-audit-exporter
      --listen-address string   listen address (default "0.0.0.0:9090")
      --metrics-path string     metrics path (default "/metrics")

Run

Run binary directly:

$ linux-audit-exporter

Or, use Docker image:

$ docker run -p 9090:9090 --privileged TeachersPayTeachers/linux-audit-exporter

Get metrics

$ curl localhost:9090/metrics | grep linux_audit

Deploy

Docker

Docker images are published here.

$ docker run --privileged teacherspayteachers/linux-audit-exporter:latest

Helm

A Helm chart is available, which deploys linux-audit-exporter as a DaemonSet.

$ helm repo add tpt https://teacherspayteachers.github.io/helm-charts
$ helm install linux-audit-exporter tpt/linux-audit-exporter

The Helm chart code is located here.

Metrics

When run from a Linux host with required privileges, will export Linux Audit status as Prometheus metrics. Here is a sample:

# HELP linux_audit_backlog Number of event records currently queued waiting for auditd to read them.
# TYPE linux_audit_backlog gauge
linux_audit_backlog 0
# HELP linux_audit_backlog_limit Number of outstanding audit buffers allowed.
# TYPE linux_audit_backlog_limit gauge
linux_audit_backlog_limit 5000
# HELP linux_audit_backlog_wait_time Time kernel waits when backlog limit is reached.
# TYPE linux_audit_backlog_wait_time gauge
linux_audit_backlog_wait_time 15000
# HELP linux_audit_backlog_wait_time_actual Total time spent by kernel waiting to queue audit events on backlog.
# TYPE linux_audit_backlog_wait_time_actual gauge
linux_audit_backlog_wait_time_actual 10
# HELP linux_audit_enabled Enabled flag. 0 = disabled. 1 = enabled. 2 = immutable. -1 = unknown.
# TYPE linux_audit_enabled gauge
linux_audit_enabled 1
# HELP linux_audit_failure Number of critical errors, such as transmission errors, backlog limit exceeded, etc.
# TYPE linux_audit_failure gauge
linux_audit_failure 0
# HELP linux_audit_lost Number of event records that have been discarded due to kernel audit queue overflowing.
# TYPE linux_audit_lost gauge
linux_audit_lost 0
# HELP linux_audit_rate_limit Limit of messages per second. A value of zero means no rate limit is applied.
# TYPE linux_audit_rate_limit gauge
linux_audit_rate_limit 100000

Alternatives

printk and dmesg

The Linux audit system can be configured to log failures and lost events with printk, which can usually be read with dmesg.

StatsD plugin

There is an (experimental, at time of this writing) user-space StatsD plugin.

Contributing

Contributions are very welcome!

Please see CONTRIBUTING.md for information on how to contribute changes to this project.

Release

When new commits are merged to the main branch, an internal process will automatically create a new GitHub release, Docker image, and Helm chart if there are any new fix: or feat: commits.

Note to Teachers Pay Teachers employees: see this internal wiki for information about this process.

Continuous Integration

New commits or pull requests to this project are automatically built, linted and tested by GitHub Actions. See the ./github/workflows/ci.yaml file for the configuration of CI actions.

License

MIT